Check Your 4090's 12vhpwr Connectors To Make Sure They are in Pristine Condition

[pc-guy]

This will happen naturally because the wires and connectors have non-zero resistance. There's even some negative feedback because copper has a positive temperature coefficient (hotter wires have more resistance, and so pass a smaller proportion of the current compared to their neighbours).

normally speaking - I have no experience on the RTX 30 series 12pin adaptor (this is the FE adaptor)

however say a RTX 3090 has 3 PCIE8 connector - lets call each of these power connector are a circuit and forget about the fact the GPU further split the power into 3/4 phases. so each circuit has its own 12V rail (i know they ultimately all connected to the same rail at PSU but for the purpose of understanding how GPU normally regulate power, lets go along with this). each of these circuit has sense cable to the PSU connecting the GPU to the PSU in terms of power demand.

so a 3090 should be able to demand different about of power from these 3 seperate circuits.

what they have done now with the adaptor seems to be putting everthing on a shared circuit which is confusing. whats the point of 4 sense cable...or even 2 in the case of the adaptor, technically you just need 1 cos all the PSU wants to know is when to delivery 600W on those PCIE8Pins

 

[arknor]

Well Igor quickly changed his mind

Now he believes the problem is user error, people incorrectly connecting the cable

Cause simpler than thought? NVIDIA GeForce RTX 4090, the scorched 12VHPWR connector and how to plug it in | Blind Test | igor´sLAB

Sometimes you also have to listen to your children and do a blind test once. Apart from the fact that there are obviously three or four adapter types with different solder pads and cables…

www.igorslab.de

yet everytime I said that I got told no that's wrong... see my last replies saying people on reddit kept posting photos shoiwing an inccorectly inserted cable, then they backtrack and say that did that after it was already damaged to show it was.... which makes 0 sense

 

[Murphy]

the new connector is 16PIN not 12PIN.

there are 12 power pins and 4 sense pins. the way it works is that the 16pin connector somehow average the load across allt he PCIE8pin connectors. I am not sure how it manages it as everything I know about power cables tells me that is such a bad idea.

Technically yes but they're called 12pin and like i said i know how it works, 12 x 12v, 12 x grounds, and 8 sense from four 8 pin PCIe power connectors get connected to the adaptor that combines each of the 8 pin PCIe power connectors into 4 single wire pairs with 4 sense wires, those 8 wires (12 if you want to include the sense wires in that) to a 12 pin connector (16 if you're including the sense wires). Like i keep saying I'm not questioning how they connect, i fully understand the how, I'm questioning why they're taking 8 wires (12 if you include the sense wires) and connecting them to 12 pins (16 if you include the sense wires) instead of taking 8 wires (12 if you include the sense wires) and connecting them to 8 pins (12 if you inclued the sense wires).

I mean the sense wires also seem pointless to me as you don't need extra wires to know if something is grounded or powered but that's an entirely different thing.

 

[pc-guy]

Technically yes but they're called 12pin and like i said i know how it works, 12 x 12v, 12 x grounds, and 8 sense from four 8 pin PCIe power connectors get connected to the adaptor that combines each of the 8 pin PCIe power connectors into 4 single wire pairs with 4 sense wires, those 8 wires (12 if you want to include the sense wires in that) to a 12 pin connector (16 if you're including the sense wires). Like i keep saying I'm not questioning how they connect, i fully understand the how, I'm questioning why they're taking 8 wires (12 if you include the sense wires) and connecting them to 12 pins (16 if you include the sense wires).

I mean the sense wires also seem pointless to me as you don't need extra wires to know if something is grounded or powered but that's an entirely different thing.

i believe in the new ATX3 PSUs everything is direct wire.

in the case of the adaptor, i think they just wanted to play on the safe side. i mean asking 1 wire to carry 150w each is a big ask - also if somehow one of the PCIE8 becomes faulty or start to drop load - it means the other 3 are carrying more so you are looking at even bigger load. I guess the 4 is for redundancy and safety - which probably made the thing more complex than it needs to be and thus causing some internal design issues.

allow only 150w on a pcie 8 pin is very conservative.

 

[Mr Evil]

...what they have done now with the adaptor seems to be putting everthing on a shared circuit which is confusing. whats the point of 4 sense cable...or even 2 in the case of the adaptor, technically you just need 1 cos all the PSU wants to know is when to delivery 600W on those PCIE8Pins

I think you might be thinking that the sense pins are for remote voltage sensing. Of the four sense pins, only two are used, and they are intended for the PSU to tell the GPU how much power it is safe to draw. Since ATX 2 PSUs can't do that, it's nVidia's adapter that sends the appropriate signal to the GPU over those pins, depending on how many cables are connected.

Gamer's Nexus have a video where they explain how the power capability detection works (I'm not going to go and find an URL to the exact video now).

Unfortunately it's not possible to view the PCIe standard docs to get the pinout of the 12VHPWR conenctor, but here's a 3rd party who show picture of the relevant part, along with a description of what the pins do: https://help.moddiy.com/en/article/...al-pins-are-used-in-the-12vhpwr-cable-yg3cv4/

So, the sense pins don't actually affect how much power is delivered over each of the four cables. With all four cables connected together at both ends, and all of them the same length and therefore the same resistance, they should share current evenly.

 

[Murphy]

in the case of the adaptor, i think they just wanted to play on the safe side.

But one wire carrying 150W is exactly what they're doing, every adaptor that's had it's braiding pulled back shows 2 wires (1 x +12v & 1x ground) coming from each 8pin PCIe connector (can we ignore the sense pins for now as they just confuse things) soldered to one or in some cases two of the 12 pin terminals.

 

[pc-guy]

makes sense why all the adaptors only have 2 sense pin connected.

the addtional 2 are completely redundant for ATX2

 

[pc-guy]

But one wire carrying 150W is exactly what they're doing, every adaptor that's had it's braiding pulled back shows 2 wires (1 x +12v & 1x ground) coming from each 8pin PCIe connector (can we ignore the sense pins for now as they just confuse things) soldered to one or in some cases two of the 12 pin terminals.

its a thicker gauge tho. 14AWG vs 18AWG

 

[Murphy]

its a thicker gauge tho. 14AWG vs 18AWG

But that doesn't change the fact that 4 pairs are going into 12 connectors.

 

[pc-guy]

4 pairs of 14AWG is pretty substantial carrying capacity.

It isn’t the best solution in all fairness.

I think the lower sku cards won’t see this craziness. Cos the TDP isn’t as high - who knows.

 

[Murphy]

Like i said though it's not the gauge of the wire, it's that 4 +12v and 4 grounds are going into 6 +12v pins and 6 ground pins, it's that i don't understand what the reason is for soldering two of the wires being soldered to 2 pins each while another two wires are soldered to individual pins. It's that i don't understand why, from an electrical/electronic POV you want to split two of the wires into 4 pins while the other two are soldered to dedicated pins.

E.g Why solder the two middle wires to two pin each when the outside wires use dedicated pins, why not just one pin for each wire.

 

[pc-guy]

Like i said though it's not the gauge of the wire, it's that 4 +12v and 4 grounds are going into 6 +12v pins and 6 ground pins, it's that i don't understand what the reason is for soldering two of the wires being soldered to 2 pins each while another two wires are soldered to individual pins. It's that i don't understand why, from an electrical/electronic POV you want to split two of the wires into 4 pins while the other two are soldered to dedicated pins.

E.g Why solder the two middle wires to two pin each when the outside wires use dedicated pins, why not just one pin for each wire.

That connector is atypical. No one has found theirs to be this arrangement yet.

It seems Igor has a unique connector. He s not come clean with where his one come from etc. it is hard to ascertain anything.

This is the BS that is about his article, it just raise more questions than it answers.

GN has done a good video, I think people just need to forget about what Igor has posted and put that aside and take a clean state of things. GN shows 2 wires (14AWG) going into 3 pins off a bus-bar.

This is also seem to be the case for someone on Reddit who opened their adaptor up. But they have 150V wires instead of the 300V from GN

 

[Murphy]

But even so you still end up with four wires connecting to 6 pins (one lot for +12v and one lot for grounds).

Here, GN video time stamp of 4 wires going to 6 pins

 

[pc-guy]

Yeah that is the design for the adaptor. I reckon they should have supplied an ATX3 PSU for that price

Then there would have been no problems

 

[Murphy]

I know it's the design. What i don't get is if there's any reason to do it that way, i mean it seems like there's two redundant pins as like I've been saying i can't see why two of the wires have dedicated pins while another two get split across two pins.

I'm not an electric or electronics engineer so maybe there's a reason for doing that that i just don't understand.

 

[EastCoastHandle]

I know it's the design. What i don't get is if there's any reason to do it that way, i mean it seems like there's two redundant pins as like I've been saying i can't see why two of the wires have dedicated pins while another two get split across two pins.

I'm not an electric or electronics engineer so maybe there's a reason for doing that that i just don't understand.

They did it this way because it's cheaper to produce verses the correct way.
Now we have a scandal of people's connectors melting. A 1st in the last few gens on mass like this.

The root cause, IMO, is that Nv made a GPU in which you need a true atx 3.0 psu with gen 5 connectors. Because people don't have such a psu and the fact that most won't upgrade to one just to buy a $1600+ gpu we have these "adapters" included in the box. Problem is these adapters haven't been tested, to my knowledge, of how power consumption is handled when starting the PC several times. In such cases power consumption can sky rocket when using atx 2.0 PSU. As there is no "checks" in place to regulate power balancing. Which is why you don't see all the sense pins in use. Which is the point of 12vhpwr to begin with but I digress.

But here we are waiting to see how many people actually report the issue on "reddit" vs anywhere else just to get a glimpse of how bad the issue really is.

 

[pc-guy]

They did it this way because it's cheaper to produce verses the correct way.
Now we have a scandal of people's connectors melting. A 1st in the last few gens on mass like this.

The root cause, IMO, is that Nv made a GPU in which you need a true atx 3.0 psu with gen 5 connectors. Because people don't have such a psu and the fact that most won't upgrade to one just to buy a $1600+ gpu we have these "adapters" included in the box. Problem is these adapters haven't been tested, to my knowledge, of how power consumption is handled when starting the PC several times. In such cases [ower consumption can sky rocket when using atx 2.0 PSU. As there is no "checks" in place to regulate power balancing. Which is why you don't see all the sense pins in use. Which is the point of 12vhpwr to begin with but I digress.

But here we are waiting to see how many people actually report the issue on "reddit" vs anywhere else just to get a glimpse of how bad the issue really is.

There are a few things there are conjectures

1) there are no more than a dozen burnt connectors
2) out of the burnt ones - it seems majority are due to user error ie not plugged in properly or severely bent
3) I do believe they have been tested otherwise they wouldn’t have put these out to the public. How rigorous the testing are that’s another question. But to say no testing done is wrong
4) I don’t think the 4090 GPU will exceed the 600w power envelop so technically the design is ok as it caters for the maximum power envelop. But it is not a smart way of doing thing nor the industrial standard. But tbh people mining have been doing this for yonks so it is not uncommon.

 

[Murphy]

They did it this way because it's cheaper to produce verses the correct way.

It's almost certainly not that, when you're making 60-70% profit nickel-and-diming on a connector simply isn't a consideration. Plus a 12 pin high power connector is probably cost a lot more to design, engineer, and make than simply up-rating and making a few minor changes to an existing 8 pin PCIe connector.

I'd hypothesises, because I'm not an electrical/electronics engineer, that the reason we've never see something like this happen before is because we've never pushed so much power through single wires/connectors. Unless I'm mistaken the most power on a +12V wire/pin is the EPS12V and that's 100W (afaik), 8pin PCIe is 50W per wire/pin, this 12VHPWR adaptor pushes 150W per wire/pin.

 

[Surveyor]

1) there are no more than a dozen burnt connectors

There are more than a dozen listed in post #444 and I doubt everyone with a problem takes it to the internet.

 

[Grim5]

Maybe too early to tell but I remember at the launch of the 2080ti, some cards were dying due to bad vram modules. The cards would typically die within 2 weeks of use. At the time, a number of forum members here experienced failure including myself.

The community thought the issue was massively widespread. After several months once all affected cards were rma'd and numbers counted, it turned out the issue affected just 1% of cards sold.

Now some people think this new issue is widespread but we don't even have one OCUK member with the issue yet.